Mean Shift Clustering

George Seif

library(ggplot2)
library(dplyr)
library(MeanShift)
library(plotly)
library(knitr)
opts_chunk$set(fig.width=9.5)

address <- url("http://www.trutschnig.net/RTR2015.RData")
load(address)
df <- RTR2015[sample(nrow(RTR2015), 10000), ]

plot_ly(df, x = ~longitude, y = ~latitude, size=~rtr_speed_dl)

No trace type specified:
  Based on info supplied, a 'scatter' trace seems appropriate.
  Read more about this trace type -> https://plot.ly/r/reference/#scatter
No scatter mode specifed:
  Setting the mode to markers
  Read more about this attribute -> https://plot.ly/r/reference/#scatter-mode
`line.width` does not currently support multiple values.No trace type specified:
  Based on info supplied, a 'scatter' trace seems appropriate.
  Read more about this trace type -> https://plot.ly/r/reference/#scatter
No scatter mode specifed:
  Setting the mode to markers
  Read more about this attribute -> https://plot.ly/r/reference/#scatter-mode
`line.width` does not currently support multiple values.

spatial_df_t <- t(df[,c("longitude", "latitude", "rtr_speed_dl")])

Apply standardization and form a set of candidate bandwiths.

spatial_df_t <- spatial_df_t / apply(spatial_df_t, 1, sd)
h.cand <- quantile(dist(t(spatial_df_t)), seq(0.05, 0.40, by=0.05))

system.time(
  bms.clustering <- lapply(h.cand, 
                           function(h){
                             bmsClustering(spatial_df_t, h=h)
                           )
  )

Running blurring mean-shift algorithm...

resulting_df_s <- as.data.frame(t(spatial_df_t)) %>% select(longitude, latitude, rtr_speed_dl) %>% mutate(cluster = as.factor(bms.clustering[[6]]$labels))
(resulting_df <- test_df %>% 
  select(longitude, latitude, rtr_speed_dl) %>% 
  mutate(cluster = as.factor(bms.clustering[[6]]$labels), rtr_speed_dl_s = rtr_speed_dl/sd(rtr_speed_dl)))

g <- list(
  scope = 'europe',
  showland = TRUE,
  # landcolor = toRGB("gray95"),
  # subunitcolor = toRGB("gray85"),
  # countrycolor = toRGB("gray85"),
  countrywidth = 1,
  subunitwidth = 1
)
plot_geo(resulting_df, lon = ~longitude, lat = ~latitude) %>% 
  add_markers(size=~rtr_speed_dl_s
              , color=~cluster
              , text=~paste(paste("Download Speed: ", rtr_speed_dl)
                            , paste("Long: ", longitude)
                            , paste("Lat: ", latitude)
                            , sep = "<br />"
                            )
              , hoverinfo="text"
              ) %>% 
  layout(
    title = 'Mobile download speed in Austria'
    , geo = g
    , mapbox = list(style = "satellite-streets")
  )

`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.`line.width` does not currently support multiple values.

plot( spatial_df_t[1,], spatial_df_t[2,], col=bms.clustering[[5]]$labels, 
xlab="longitude", ylab="latitude", main="Mean shift labels",
cex=spatial_df_t[3,], pch=16 )
points( bms.clustering[[5]]$components[1,], bms.clustering[[5]]$components[2,], col=1:ncol( bms.clustering[[5]]$components ),
pch="+", cex=3 )

p <- plot_ly(spatial_df_t %>% 
               t() %>% 
               as.data.frame()
             , x = ~longitude
             , y = ~latitude
             , z = ~rtr_speed_dl) %>%
  add_markers(color = bms.clustering[[5]]$labels)
p